Evaluation of fuzzy learning-based scaling policies for the Slingshot Simulator

Evaluation results for Fuzzy Learning models in the Slingshot simulator

This dataset contains the evaluation results for the evaluation of the Fuzzy Learning implementation in the Slingshot simulator (See https://github.com/PalladioSimulator/Palladio-Addons-SPD-Metamodel/pull/32, https://github.com/PalladioSimulator/Palladio-Analyzer-Slingshot-Extension-SPD-Interpreter/pull/49).

For the evaluation, the evalution of the paper proposing these Fuzzy Learning methods [^1] was replicated in Slingshot. The models used for replication are contained in `PCM + SPD Models.zip`.

The different zip files contain the raw output of the Slingshot simulator for different simulation runs.

The folder names refer to the experiment, i.e. `reactive_load_45-80` contain the evaluations for a reactive scaling policy that scales in at 45% CPU load and scales out at 80% CPU load. The folders named like `reward_0_0_thrashing_pessimistic` contain the evaluations for learning-based policies, specifying the reward given to the model for scaling in and out and the initialization method, such as `pessimistic`. Folders postfixed with `_90p` contain evaluations for which the model has a 90th-percentile target policy.

The zipped folders additionally contain some processed data such as graphs showing the response times of the runs or how many containers a model allocated.

`analysis.csv` contains some calculated values such as the average container count or share of SLO violations for each of the containted evaluations.

For completeness, the python and jupyter notebook files used to generate these evaluations are also included.  The two `.patch` files represent additional patches to the Slingshot simulator components that can be used to replicate the reduced logging intensity used for the evalautions. The `.launch` files are the launch configurations used for the evaluation of the different workload patterns.

[^1]: H. Arabnejad, C. Pahl, P. Jamshidi, and G. Estrada, “A Comparison of Reinforcement Learning Techniques for Fuzzy Cloud Auto-Scaling,” May 19, 2017, arXiv: arXiv:1705.07114. Accessed: Jun. 13, 2024. [Online]. Available: http://arxiv.org/abs/1705.07114